The same poll, two defensible numbers

The non-response group is larger than any candidate's support. At 82 people it is bigger than Plaskett's 56, and bigger than Frett-Gregory's 27 and Roach's 15 combined. Rattan does not describe why those 82 did not answer, and there is a real difference between a refusal, an undecided voter and a skipped question that this page cannot resolve from the published sheet. What can be said is that the 2022 poll placed that group inside its percentages and the 2026 sheet does not, so the two years cannot be read side by side without this adjustment. VIIQ arithmetic from Rattan's published raw counts.

The published margin of error is right, and it applies to nothing

Rattan states a margin of error of plus or minus 7.3 points at 95% confidence. That figure is correct for a sample of 180. It checks exactly: 1.96 multiplied by the square root of 0.25 divided by 180 gives 7.30%. The difficulty is that no horse-race question on this poll was answered by 180 people, and margin of error scales to the number who answered, not the number who were asked.

All margins calculated at the 95% confidence level with the conservative assumption of a 50% response share, which is the same assumption behind Rattan's own 7.3%. Formula: 1.96 multiplied by the square root of 0.25 divided by n. The two questions every respondent answered carry the stated margin correctly. The three questions people will quote do not. The Senator-at-Large figure, which two thirds of the sample skipped, carries a margin nearly twice the one printed on the sheet. VIIQ calculation.

Governor: the lead survives the correct margin

Even at the wider and correct margin, Plaskett's floor sits above Frett-Gregory's ceiling. Tested properly, as a difference between two shares of the same 98 respondents, the 29.6-point gap carries a 95% interval running from 12.3 to 46.8 points. It does not cross zero. Within St. Croix, and within this sample, that lead is real.

Bands are individual 95% intervals on a base of 98. Plaskett's lower bound of 47.3% sits 10.9 points above Frett-Gregory's upper bound of 36.4%, so the two ranges do not touch. VIIQ calculation.

Delegate: the lead does not survive it

Sarauw leads Smith by 7.4 points on a base of 94. Tested the same way, that gap carries a 95% interval of minus 8.2 to plus 23.1 points. It crosses zero comfortably. So does Sarauw's lead over the write-in group, at minus 7.0 to plus 24.0. On this poll the delegate race has no measurable leader. The top three are statistically indistinguishable from one another, and one of the three is not a candidate.

The top three bands overlap across almost their entire length. Any of Sarauw, Smith or the write-in group could be first on these figures. Note also that a quarter of everyone who answered named someone who was not on Rattan's list, which is the third-largest response in the race and the single most under-discussed number on the sheet. VIIQ calculation on a base of 94.

Rattan's record, scored against certified results

Rattan Polling Service states it has served the territory since 1998. VIIQ has been able to locate one Rattan poll with a published result that can be checked against a certified election, from the 2022 Democratic primary, plus Rattan's own February to March 2026 territory-wide poll, whose outcome is still pending. No Rattan poll for 2014 or 2018 appears in any source VIIQ has searched. That is a gap in what is publicly available, and not a statement that no such poll was conducted.

Only one row in this table can be graded. The 2022 Rattan poll carries no published field dates and no published margin of error in any source VIIQ has located, which limits how it can be read: a poll taken in June and a poll taken the week before an election are not the same instrument. VIIQ has asked Rattan for the original release. If you have a Rattan poll release from 2014, 2018 or any earlier cycle, in any format, write to signal@viiq.org.

2022: right territory-wide, wrong on St. Croix

Rattan's 2022 poll reported large undecided groups, so the fair way to score it is to allocate those proportionally and compare the two-way share against the certified two-way share. Done that way, Rattan was within about six points in all three geographies, and it still called the wrong leader in St. Croix.

Certified figures: territory-wide Bryan 4,269 and Bernier 2,255; St. Croix Bryan 1,656 and Bernier 1,432; St. Thomas and St. John Bryan 2,613 and Bernier 823. Error is the poll's two-way share of the eventual winner minus the certified two-way share. Mean absolute error across the three geographies was 5.7 points, which for a sample of 160 is close to what sampling alone would produce. The margin misses were larger and unusually consistent, at 10.9, 12.0 and 11.1 points. VIIQ calculation from Rattan's published figures and Elections System certified results.

February to July, inside Rattan's own numbers

Rattan polled the territory in February and March with a sample of 500 and a hypothetical field of thirteen names. Three of those names are the actual ticket heads on Saturday's ballot. Rattan published how each candidate's support divided between the islands, which allows a St. Croix figure to be reconstructed from the earlier poll and set beside the July one.

Restricted to the three real candidates and to St. Croix, Plaskett is roughly flat, Roach has lost more than a third of his share, and Frett-Gregory has nearly doubled hers. Frett-Gregory and Roach have swapped second place on St. Croix in four months.

How the February figures were reconstructed, and why they are shown with caution. Rattan's February poll reported that Plaskett's support divided 50 / 50 between St. Croix and St. Thomas, Roach's 40 / 60, and Frett-Gregory's 30 / 70. Applying those splits to their reported respondent counts of 94, 47 and 40 gives St. Croix support of roughly 47, 19 and 12, a three-way base of about 78. That base is smaller than July's 98, and it is VIIQ arithmetic on Rattan's percentages rather than a figure Rattan published. The two polls also differ in field, in geography and in whether Kurt Vialet was among the names offered, and he was the leader in February. Treat the direction of movement as the finding and the size of it as approximate.

One district is not the territory

The most important limitation on this poll is the one printed on its own masthead. It surveyed the St. Croix District only. The governor's race is decided territory-wide, and St. Croix produced 3,461 of the 7,113 early ballots cast this year, or 48.7%. The other 51.3% of the early electorate was not asked anything.

That matters more than usual here because the field is geographically split. Plaskett is the St. Croix candidate. Frett-Gregory is a former St. Thomas senator. Roach is the sitting Lieutenant Governor and, on Rattan's own February splits, drew 60% of his support from St. Thomas. A St. Croix-only poll is the most favourable ground any of these three could be measured on, for exactly one of them.

This is arithmetic, not a forecast, and it is printed so readers can do the comparison rather than guess at it. It assumes Saturday reproduces the early-voting district split of 48.7% St. Croix and 51.3% St. Thomas and St. John, which it need not, and it assumes Rattan's St. Croix figures hold, which is the thing being tested. What the table shows is the shape of the problem: on these numbers Plaskett would need a little under a quarter of the other district to reach 40% territory-wide, while Frett-Gregory would need better than half of it and Roach better than three fifths. No poll of the St. Thomas and St. John district for this race has been located, by Rattan or anyone else. That is the missing half of the picture, and it is missing entirely. VIIQ calculation from Rattan's July figures and Elections System early voting totals.

The Senate read, and the one place Rattan has been consistent

The St. Croix Senate question is the strongest part of this poll and the part least likely to be misread, because every respondent answered it. It is also the one race where Rattan's order can be checked against a recent certified result. Its top four match the 2024 certified order exactly, once the one departing name is removed.

Remove Novelle Francis, who is running for Lieutenant Governor in 2026 rather than for the Senate, and the 2024 certified order of the returning names reads Vialet, Joseph, Gittens, Frederick. That is Rattan's 2026 top four in the same sequence. Six of the ten names Rattan polled did not appear in the 2024 St. Croix Senate primary and have no certified rank to check against, so this is a test of four names rather than ten, and it is a test of a poll against a race held two years earlier under different conditions. It is also, in fairness, largely a measure of durable name recognition among incumbents and near-incumbents. 2024 figures from the certified St. Croix Senate primary result.

What voters said matters, and what changed

Every respondent answered the issues question, so it carries the stated margin correctly and is the most robust part of the poll. Comparing it to February requires care: the February question appears to have asked for a single most important issue while July allowed several, so the levels are not comparable. The ordering is.

Two of the three largest moves are probably about the question rather than the voters. WAPA and energy did not appear as an option in February at all, so its arrival at second place cannot be read as a rise in concern. Corruption in February and government accountability in July are related but not identical wordings, and the drop from fourth to last may partly reflect that. The economy falling from first territory-wide to seventh on St. Croix is the move least explained by wording, since the option was offered both times under the same name, though geography and question format both changed. Fifteen respondents separate the top issue from the bottom three, which on a base of 180 is inside the margin. Treat the top group and the bottom group as distinguishable and the ordering within each as noise.

Discrepancies on the record

VIIQ checks published figures against their own arithmetic before analysing them. The July sheet is internally consistent almost everywhere: every reported percentage divides correctly, the Senate selections sum to 317, the issue responses sum to 233, and the three race subtotals of 98, 94 and 60 each reconcile with their stated non-response counts. The items below are the exceptions, and they are recorded rather than corrected.

• The summary card and the full sheet disagree. The one-page results card gives Roach 15.5%; the full sheet gives 15.3%, which is what 15 divided by 98 produces. The card also lists Lisa Charles ahead of Jose Gonzalez and Nemmy Williams-Jackson at a flat 17% each, while the sheet gives Gonzalez and Williams-Jackson 32 selections and Charles 31. A one-selection difference, and the card is the version that circulates.
• The sample is described two ways. The sheet is headed "180 registered Democratic voters" and separately reports that 85% of respondents identified as Democrats. Both can be true, since party registration and self-identification are different things and the 15% gap is itself a finding worth having, but presented together without explanation they read as a contradiction.
• Five ballots are unaccounted for. The methodology states that five ballots could not be accounted for and were excluded. It is not stated whether the sample was therefore 180 or 185. Every percentage on the sheet divides cleanly by 180, so 180 is used throughout this page.
• The February poll does not fully reconcile. Its reported respondent counts sum to 466 of 500 and its percentages to 93.2%, leaving 34 respondents unaccounted for beyond the 3.0% recorded as undecided. Two reported counts also do not match their own percentages: education is given as 8.6% and 48 respondents, where 8.6% of 500 is 43, and restarting the refinery as 8.0% and 43 respondents, where 8.0% of 500 is 40. The pattern suggests a transposition somewhere in reporting rather than an error of substance.
• The 2022 Rattan poll has no published field dates or margin of error in any source VIIQ has located. A poll taken in June and a poll taken the week before an election are different instruments, and without dates the 2022 accuracy scoring above should be read as indicative rather than exact.

What this poll does and doesn't tell you

• The denominator is the story. 57.1% is 56 people out of 98 who answered, inside 180 surveyed. On the basis Rattan itself used in 2022, with non-responders left in, the figure is 31.1%. Both are correct; only one is comparable across years, and it is not the one on the graphic.
• The stated margin of error understates every race on the sheet. Plus or minus 7.3 points is right for 180 respondents and applies only to the issues and Senate questions. Governor is plus or minus 9.9, delegate plus or minus 10.1, Senator-at-Large plus or minus 12.7.
• Plaskett's St. Croix lead is real, on this sample. Tested as a difference between two shares of the same 98 respondents, the 29.6-point gap over Frett-Gregory carries a 95% interval of 12.3 to 46.8 points and does not approach zero. Widening the margin correctly does not rescue second place.
• The delegate race has no measurable leader. Sarauw's 7.4-point lead over Smith carries an interval of minus 8.2 to plus 23.1. Her lead over the write-in group is no firmer. On this poll the top three are indistinguishable, and one of the three is not a candidate.
• A quarter of delegate answers named nobody on the list. "Other or write-in" at 25.5% is the third-largest response in that race and is not being discussed. Given the unresolved litigation over candidate disqualifications, it is a figure worth understanding before it is explained away.
• Rattan's district calls have been shakier than its territory-wide call. In 2022 it was within 5.4 points territory-wide and called the winner correctly, and in St. Croix it was 6.0 points off and called the wrong leader. The 2026 poll is district-level only. That is one prior cycle and this page does not turn it into an adjustment.
• Half the electorate was not surveyed. St. Croix cast 48.7% of this year's early ballots. No poll of the St. Thomas and St. John district has been located for this race, from Rattan or anyone else, and the governor's race is decided territory-wide.
• The geography of the field cuts against reading this as a territorial number. Plaskett is the St. Croix candidate, Frett-Gregory a former St. Thomas senator, and Roach drew 60% of his February support from St. Thomas on Rattan's own reporting. A St. Croix-only poll is favourable ground for exactly one of the three.
• Frett-Gregory has moved on St. Croix and Roach has fallen. Reconstructed from Rattan's February district splits, Frett-Gregory has gone from about 15% to 28% of the three-way there and Roach from about 24% to 15%. The reconstruction rests on a base of roughly 78, so the direction is more reliable than the size.
• The Senate ordering has been stable across two years. Rattan's top four on St. Croix are the same four names, in the same order, as the 2024 certified result once Novelle Francis is removed. That is a real consistency signal and also, largely, a measure of name recognition among sitting and former officeholders.
• Senate percentages cannot be added or turned into seats. Voters could name seven and named 1.76 on average, so the figures sum to 176% and measure recall rather than a filled ballot. Rattan does not publish the overlap between selections, so the gaps between candidates cannot be tested for significance at all.
• Stated turnout intent is far above observed behaviour. 70% said they will definitely vote, during a field period that overlapped early voting almost exactly. The territory-wide early vote reached 34% of active registered Democrats, and in 2022 the Election Day and absentee period added fewer ballots than the early period did.
• Nothing here is a projection, including the threshold table. The territory-wide figures in this analysis show what Saturday would have to produce for these poll numbers to translate into a given result. They are subtractions from published totals. VIIQ has not modelled, projected or forecast a result on this page and does not intend to.

Sources and method

Poll figures come from Rattan Polling Service's published results sheet and summary card for the St. Croix District, from Virgin Islands Consortium reporting of Rattan's February to March 2026 territory-wide poll, and from the Rattan poll recorded in the public election archive for 2022. Election results come from Elections System of the Virgin Islands certified reports. Margins of error, confidence intervals, significance tests, two-way allocations, the February St. Croix reconstruction, the district thresholds and every basis conversion are VIIQ calculations from those figures and are labelled as such throughout. No figure on this page is estimated, projected or modelled. Where a figure has not been published, this page says so rather than filling the gap.